A Proteomic Network Approach Across the ALS-FTD Disease Spectrum

A proteomic network approach across the ALS-FTD disease spectrum resolves clinical phenotypes and genetic vulnerability in human brain Mfon E Umoh, Emory University School of Medicine Eric Dammer, Emory University Jingting Dai, Emory University School of Medicine Duc M Duong, Emory University School of Medicine James Lah, Emory University Allan Levey, Emory University Marla Gearing, Emory University Jonathan Glass, Emory University Nicholas Seyfried, Emory University

Journal Title: EMBO Molecular Medicine Publisher: Wiley Open Access | 2017-01-01 Type of Work: Article | Final Publisher PDF Publisher DOI: 10.15252/emmm.201708202 Permanent URL: https://pid.emory.edu/ark:/25593/s730k

Final published version: http://dx.doi.org/10.15252/emmm.201708202 Copyright information: © 2017 EMBO. This is an Open Access work distributed under the terms of the Creative Commons Attribution 4.0 International License (https://creativecommons.org/licenses/by/4.0/). Accessed September 30, 2021 4:20 AM EDT Research Article

A proteomic network approach across the ALS-FTD disease spectrum resolves clinical phenotypes and genetic vulnerability in human brain

Mfon E Umoh1,2,†, Eric B Dammer2,3,†, Jingting Dai2,3, Duc M Duong2,3, James J Lah1,2, Allan I Levey1,2, Marla Gearing1,2,4, Jonathan D Glass1,2,4,* & Nicholas T Seyfried1,2,3,**

Abstract Subject Categories Neuroscience; Post-translational Modifications, Proteoly- sis & Proteomics Amyotrophic lateral sclerosis (ALS) and frontotemporal dementia DOI 10.15252/emmm.201708202 | Received 27 June 2017 | Revised 13 October (FTD) are neurodegenerative diseases with overlap in clinical 2017 | Accepted 20 October 2017 | Published online 30 November 2017 presentation, neuropathology, and genetic underpinnings. The EMBO Mol Med (2018) 10: 48–62 molecular basis for the overlap of these disorders is not well established. We performed a comparative unbiased mass spectrometry- based proteomic analysis of frontal cortical tissues from post- Introduction mortem cases clinically defined as ALS, FTD, ALS and FTD (ALS/FTD), and controls. We also included a subset of patients with the Amyotrophic lateral sclerosis (ALS) and frontotemporal dementia C9orf72 expansion mutation, the most common genetic cause of (FTD) are clinically distinct neurodegenerative diseases that are both ALS and FTD. Our systems-level analysis of the brain connected by genetic and pathological overlap (Fecto & Siddique, proteome integrated both differential expression and co-expres- 2011; Ferrari et al, 2011; Achi & Rudnicki, 2012). ALS patients sion approaches to assess the relationship of these differences to present with muscle weakness and spasticity associated with clinical and pathological phenotypes. Weighted co-expression degeneration of motor neurons in the motor cortex, brainstem, network analysis revealed 15 modules of co-expressed proteins, and spinal cord that ultimately leads to death. In contrast, eight of which were significantly different across the ALS-FTD patients with FTD display cognitive dysfunction associated with disease spectrum. These included modules associated with RNA degeneration of neurons in the frontal and temporal lobes of the binding proteins, synaptic transmission, and inflammation with brain. Despite being clinically distinct, 15% of individuals cell-type specificity that showed correlation with TDP-43 pathol- presenting with FTD also have ALS, whereas 30% of individuals ogy and cognitive dysfunction. Modules were also examined for with ALS will develop FTD (Lomen-Hoerth, 2011). This implies their overlap with TDP-43 protein–protein interactions, revealing that these two neurodegenerative diseases are part of a shared one module enriched with RNA-binding proteins and other causal clinical spectrum. ALS genes that increased in FTD/ALS and FTD cases. A module In addition to their clinical overlap, most cases of ALS and enriched with astrocyte and microglia proteins was significantly FTD display pathological accumulation of TAR-DNA binding increased in ALS cases carrying the C9orf72 mutation compared to protein (TDP-43), a ubiquitously expressed nuclear DNA/RNA sporadic ALS cases, suggesting that the genetic expansion is asso- binding protein that is cleaved, phosphorylated, and aggregated in ciated with inflammation in the brain even without clinical the cytoplasm in disease (Neumann et al, 2006). Ninety-seven evidence of dementia. Together, these findings highlight the utility percent of ALS cases display phosphorylated TDP-43 pathology in of integrative systems-level proteomic approaches to resolve clini- the brain and/or spinal cord, while 50% of FTD cases display cal phenotypes and genetic mechanisms underlying the ALS-FTD this pathology throughout the brain (Radford et al, 2015), defin- disease spectrum in human brain. ing these diseases as TDP proteinopathies. Many individuals with ALS and FTD also share a positive family history of disease (Fong Keywords C9orf72 expansion mutation; mass spectrometry; neurodegenera- et al, 2012). The largest proportion of inherited cases (40% ALS

tion; protein co-expression; TDP-43 and 25% FTD) are caused by hexanucleotide G4C2 repeat

1 Department of Neurology, Emory University School of Medicine, Atlanta, GA, USA 2 Center for Neurodegenerative Diseases, Emory University School of Medicine, Atlanta, GA, USA 3 Department of Biochemistry, Emory University School of Medicine, Atlanta, GA, USA 4 Department of Pathology and Laboratory Medicine, Emory University School of Medicine, Atlanta, GA, USA *Corresponding author. Tel: +1 404 727 3275; E-mail: [email protected] **Corresponding author. Tel: +1 404 712 9783; E-mail: [email protected] †These authors contributed equally to this work

48 EMBO Molecular Medicine Vol 10 |No1 | 2018 ª 2017 The Authors. Published under the terms of the CC BY 4.0 license Mfon E Umoh et al ALS-FTD proteomic network EMBO Molecular Medicine

expansions in the C9orf72 gene, which notably were identified Results from families with co-occurring ALS and FTD (ALS/FTD) (DeJesus-Hernandez et al, 2011; Renton et al, 2011; Majounie Proteomic signature of human brain classifies cases by et al, 2012; Radford et al, 2015). Although the function of the clinical phenotypes C9orf72 protein is not yet known, there are several theories regarding how the C9orf72 mutation leads to ALS and FTD This study offers an in-depth analysis of protein changes in the (Zhang et al, 2012; Farg et al, 2014). For example, loss of frontal cortex of patients across the ALS-FTD disease spectrum, C9orf72 protein expression is thought to inhibit autophagy and which resulted in the final quantification of 2,612 protein groups promote neuroinflammation (Farg et al, 2014; Webster et al, mapping to 2,536 unique gene symbols across all samples (Datasets 2016), whereas expression of C9orf72 gene products could cause EV1 and EV2). Our goal was to compare ALS patients with and toxicity via nuclear sense and antisense repeat-containing RNAs without clinical dementia, so our experimental case selection that sequester RNA binding proteins (abnormal RNA metabolism) grouped samples based on clinical phenotype. As FTD is clinically or by non-canonical repeat-associated non-ATG (RAN) translation heterogeneous, we limited that group to include only individuals of dipeptide repeat proteins that aggregate and block nuclear with the pathological diagnosis of frontotemporal lobar degenera- pores (Gitler & Tsuiji, 2016). Notably, this genetic mutation tion with TDP-43 inclusions (FTLD-TDP), which accounts for the bridges these two diseases that clinical and pathological overlaps majority of overlap with ALS and FTD (Neumann et al, 2009). The have previously connected. However, the underlying molecular frontal cortex was chosen because it was a likely site for discrimi- basis of the ALS-FTD clinical and pathological spectrum is not nating proteomic differences in ALS patients with and without well established. It is also unclear why patients with the same dementia, and represents an area in which TDP-43 pathological C9orf72 genetic expansion get either or both of these disparate burden has been mapped in FTD patients (Brettschneider et al, diseases. Using our knowledge of the shared clinical, pathological, 2014). In our analytic pipeline, protein expression values are and genetic features characterizing ALS and FTD, a systems-level adjusted for age, sex, and postmortem interval (PMI) covariance proteomic analysis of both sporadic and genetic (C9orf72) cases prior to downstream differential or co-expression analyses comprising the ALS-FTD spectrum was conducted to determine (Fig EV1). Differential expression by ANOVA comparisons of each common and distinct pathways that contribute to the onset and of the groups yielded subsets of significantly altered proteins across development of dementia. controls and disease groups (Dataset EV2). The number of signifi- Co-expression network analysis has been used to define modules cantly altered proteins increased sequentially when comparing of co-expressed genes or proteins linked to specific cell types, orga- controls to ALS, ALS-FTD, and FTD, respectively, which is likely nelles, and biological pathways (Miller et al, 2008; Oldham, 2014; due to the presence of clinical dementia and an increase in associ- Seyfried et al, 2017). Assessing co-expression of proteins within ated pathological burden in the frontal cortex (Fig 1A). For super- samples and relating co-expression modules to clinical and patho- vised clustering analysis, we selected 165 proteins that had at least logical endophenotypes can be defined utilizing weighted co-expres- two ANOVA Tukey pairwise comparisons of high significance sion network analysis (WGCNA), where the most centrally (P < 0.01) among the six possible comparisons across the clinical connected proteins in a module often act as key drivers (Zhang & groups. The over-represented gene ontology (GO) terms within Horvath, 2005; Oldham et al, 2008). We recently reported the first these 165 information rich proteins included cytoplasmic vesicle large-scale proteomic and transcriptomic multinetwork analysis in part, clathrin coat, and plasma membrane (Dataset EV3). Using human postmortem brain from both asymptomatic and symp- these differentially expressed protein signatures of disease, multidi- tomatic Alzheimer’s disease (AD) (Seyfried et al, 2017). This work mensional scaling (MDS) was used to stratify the relatedness of indi- revealed that several protein-driven processes related to cognitive vidual cases, which revealed that clinical phenotypes indeed decline are distinct from networks in human AD transcriptome. associate with proteomic signatures, as samples segregated into Moreover, analysis of the proteome is particularly relevant since clusters representing clinical groups (Fig 1B). Also, ALS/FTD cases neurodegenerative diseases are collectively viewed as proteinopa- distributed between the ALS and the FTD clusters, supporting an thies defined by their association with the aggregation and accumu- underlying molecular and biological spectrum designated by protein lation of misfolded proteins (Golde et al, 2013). expression that defines the clinical spectrum. Thus, this analysis Here, we report the first unbiased proteomic analysis of post- highlights that protein signatures of ALS-FTD clinical phenotypes mortem cortical tissues from clinically characterized ALS, FTD, are related to differences in molecular pathways in the frontal ALS/FTD, and healthy disease controls. A subset of C9orf72 cortex. expansion-positive (C9Pos) cases was also included. WGCNA revealed 15 modules of co-expressed proteins, eight of which Protein co-expression network analysis were significantly different across the ALS-FTD disease spectrum. These included modules associated with RNA binding proteins, Protein co-expression networks reflect relationships between protein synaptic transmission, inflammation, and cell-type specificity pathways, cell types, and physically interacting proteins within (neuronal, microglial, and astrocytic) that showed strong correla- modules. Network analyses revealed 15 modules of strongly co- tion with TDP-43 pathology and cognitive dysfunction. Compared expressed groups of proteins (Fig 2 and Dataset EV2). Each module to sporadic ALS patients, C9Pos ALS cases showed increased is defined by an eigenprotein, the most representative weighted levels for a protein module associated with astrocytic and micro- protein expression pattern across samples for a group of co- glial markers, which supports a hypothesis that links the C9orf72 expressed proteins (Seyfried et al, 2017). Eight of these 15 modules, mutation to neuroinflammation. identified by numbers that correspond to a color, ordered 1–15, with

ª 2017 The Authors EMBO Molecular Medicine Vol 10 |No1 | 2018 49 EMBO Molecular Medicine ALS-FTD proteomic network Mfon E Umoh et al

A B

ALS ALS/FTD

2 0.4 Control 5 24 0.3 ALS 11 0.2 ALS/FTD 0.1 FTD

1 25 0.0

-0.1

Dimension 3 0.4 -0.2 0.3 374 0.2 -0.3 0.1 0.0 -0.1 -0.4 -0.2 FTD -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Dimension 2 Dimension 1

Figure 1. Segregation of ALS and FTD clinical phenotypes by proteomic signatures. A Venn diagram showing a total of 442 unique proteins determined to be significantly altered by ANOVA followed by Tukey’s post hoc test (P ≤ 0.01) among three pairwise comparisons (i) ALS, (ii) ALS/FTD, and (iii) FTD versus control cases. B Supervised hierarchical clustering of 165 significant proteins altered in the frontal cortex across clinical phenotypes (Control, ALS, ALS-FTD, and FTD) was used as input for multidimensional scaling (MDS) analysis. Each dimension on the plot explains a larger proportion of variance of the dataset. Clinical groups are indicated by colors (orange—control, light blue—ALS, dark blue—FTD/ALS, and pink—FTD). Segregation of cases based on clinical grouping is indicated by colored ovals.

M1 (largest, 548 proteins) to M15 (smallest, 21 proteins), showed 2017) (Fig 2A). The remaining three significant modules (M1, M3, different patterns of co-expression across the four clinical groups. and M8) showed a decrease in protein expression in FTD cases As expected, functional annotation of frontal cortex modules classi- compared to control and ALS cases; M1 and M8 were enriched in fied the proteome with specific gene ontologies and brain cell types synaptic and neuronal proteins, while M3 was enriched with mito- among other biological sources of co-expression (Gaiteri et al, chondrial proteins (Fig 2A and Dataset EV3). Moreover, several 2014). Of the significant modules, four (M2, M6, M9, and M10) modules showed significant correlation with clinical and pathologi- showed increased expression in FTD cases compared to ALS and cal sample traits. Pathological sample traits characterize TDP-43 control cases (Fig 2A). These modules were enriched for RNA splic- pathology, which is represented by the abundance of phosphory- ing (M2), response to biotic stimuli (M6), zinc ion binding (M9), lated TDP-43 cytoplasmic inclusions in the frontal cortex (defined and homeostatic processes (M10) (Dataset EV3). One of the signifi- by the pTDP score) or label-free protein quantification (LFQ) of cant modules, M15, enriched for blood microparticles and circulat- TDP-43 by mass spectrometry (Fig EV2); for example, M5, a module ing immunoglobulin complexes, showed increased protein enriched with extracellular matrix and astrocyte proteins correlated expression in all disease groups compared to controls, consistent closely to clinical and pathological traits (Fig 3A), M14, a module with a common mechanism of blood–brain barrier breakdown in enriched with microtubule proteins and displaying dynactin as a neurodegenerative diseases (Carvey et al, 2009; Seyfried et al, hub protein, and optineurin, genes previously implicated in ALS

Figure 2. Integrated protein co-expression and differential expression across the ALS-FTD disease continuum. ▸ A Eigenproteins, which correspond to the first principal component of a given module and serve as a summary expression profile for all proteins within a module, are shown for 10 of the 15 modules generated by WGCNA. Box plots depict mean (horizontal bars) and variance (25th to 75th percentiles), with whiskers extending to the last non-outlier measurements, are shown for all four groups (control, ALS, ALS/FTD, and FTD), and representative Gene Ontology (GO) terms are listed above each eigenprotein. Hub proteins for each of these modules are also highlighted. Significance was determined by comparisons of quantified proteins within individual cases across the clinical groups to the module eigenprotein using one-way nonparametric ANOVA Kruskal–Wallis with P-values listed above plots. Outlier cases are shown as open circles beyond the error bars. B Stacked bar plots represent analysis of differential expression (pairwise comparison between disease group and control) and enrichment of differentially expressed proteins within co-expression modules. Modules are listed along the x-axis, and the height of the bar along the y-axis indicates the proportion of differentially expressed module members, while the color indicates the fold change (red is increased, and blue is decreased) according to the scale shown.

50 EMBO Molecular Medicine Vol 10 |No1 | 2018 ª 2017 The Authors Mfon E Umoh et al ALS-FTD proteomic network EMBO Molecular Medicine

A Antigen binding Zinc ion binding M15 p = 0.0043M9 p = 0.0067 0.3 0.3

0.1 0.1

-0.1 -0.1

Eigenprotein -0.3 -0.2

ALS FTD ALS FTD Control Control ALS/FTD ALS/FTD Hub Proteins Hub Proteins

HPX FGG IGHG3 PEA15 MAPK1 KPNB1 FGB IGHG4 IGLC3 CPNE6 PAFAH1B GLOD4 Response to RNA splicing Homeostatic process biotic stimulus Extracellular matrix

M2 p = 0.0046M10 p = 1.5e-05 M6 p = 0.00054 M5 p = 0.22 0.4 0.4 0.4 0.3 0.2 0.2 0.2 0.2

0.0 0.0 0.0 0.0 Eigenprotein -0.2 -0.2 -0.2 -0.2

ALS FTD ALS FTD ALS FTD ALS FTD Control Control Control Control ALS/FTD ALS/FTD ALS/FTD ALS/FTD Hub Proteins Hub Proteins Hub Proteins Hub Proteins

HIST1H2BM LMNA COTL1 SELENBP1 DDAH1 CTSD HEPACAM MSN PLEC VCL VIM ANXA2 EEF1A1 HMGB1 CSRP1 CROCC CYB5R3 ASS1 SEPT2 PRDX6 EZR FLNA PTRF COL6A1

MitochondrionSynapse Neuron differentiation Microtubule

M3 p = 0.0048M1 p = 0.0038M8 p = 0.0055 M14 p = 0.79 0.3 0.2 0.2 0.2 0.1 0.0 0.0 0.0 -0.1 -0.2 -0.2 -0.2

Eigenprotein -0.3 -0.4 -0.4 -0.4

TD ALS FTD ALS FTD ALS F ntrol ALS FTD Control Control S/FTD Control Co ALS/FTD AL ALS/FTD ALS/FTD Hub Proteins Hub Proteins Hub Proteins Hub Proteins

NDUFV1 NDUFS1 UQCRC1 AP2B1 ATP6V0A1 SYT1 MAP1B MAP1A RTN4 RUFY3 TUBB4B DCTN1 NDUFA2 NDUFA8 ATP5A1 ATP6V1B2DMXL2 DLG4 ENSA DMTN HSPA12A TUBA1A WDR44 TNR

B ALS ALS/FTD FTD Avg. Module Fold Change log 0.8 2

1.0 (Disease-control) 0.6

0.5 0.4

0.0 0.2

-0.5

Differentially Expressed 0.0 Fraction Module Members 7 11 15 9 2 10 6 5 13 3 1 12 4 14 8 7 11 15 9 2 10 6 5 13 3 1 12 4 14 8 7 11 15 9 2 10 6 5 13 3 1 12 4 14 8 Protein Modules

Figure 2.

ª 2017 The Authors EMBO Molecular Medicine Vol 10 |No1 | 2018 51 EMBO Molecular Medicine ALS-FTD proteomic network Mfon E Umoh et al

A Eigenprotein Network 1.60 1.40 1.20 0.80 0.60 Height 0.40 0.20 0.00 M7 M11 M15 M9 M2 M10 M6 M5 M13 M3 M1 M12 M4 M14 M8

1 Correlation R-value

Grouping 0.35 0.43 0.48 0.63 0.51 0.32 -0.46 -0.43 0.33 -0.46 .001) 0.5 (0.02) (0.002) (7e-04) (2e-06) (3e-04) (0.03) (0 (0.003) (0.02) (0.001)

37 0.58 0.52 -0.3 0 pTDP43 Rating 0.34 0.36 0.36 0.55 -0. -0.2805) -0.52 (0.02) (0.01) (0.01) (7e-05) (2e-05) (2e-04) (0.01) (0.04) (0. (2e-04) −0.5 TDP43 LFQ 0.34 0.43 0.59 0.5 0.61 0.31 -0.4 -0.5 -0.36 -0.51 (0.02) (0.04) (0.01) (0.003) (2e-05) (4e-04) (6e-06) (0.005) (4e-04) (3e-04) −1

B Cell Type Enrichment

3 −log(p), BH Corrected Microglia 0.00086 0.056 ***

0.098 6.1e−05 0.00086 0.011 2 Astrocyte *** *** * 0.00018 0.016 Neuron *** * 1 0.0018 1.4e−05 Oligodendrocyte *** *** 0

Figure 3. Protein co-expression organizes the proteome into modules associated with brain cell types and clinicopathological traits. A WGCNA cluster dendrogram grouped proteins (n = 2,613) measured across the frontal cortex into distinct protein modules (M1–M15) defined by dendrogram branch cutting. Modules are clustered based on relatedness defined by correlation of protein co-expression eigenproteins (indicated by position in color bar). Listed in the heatmap are bicor correlations and P-values defining relationship between module eigenprotein expression and clinical grouping (defined as 0-control, 1-ALS, 2-ALS/ FTD, and 3-FTD), pTDP-43 rating (defined as a score from 0 to 3 that represents pathological TDP-43 burden in the frontal cortex), and TDP-43 label-free quantification (LFQ). B Cell-type enrichment analysis was performed using a one-tailed Fisher’s exact test against lists of proteins previously generated from acutely isolated neurons, oligodendrocytes, astrocytes, and microglia. The heatmap displays Benjamin–Hochberg-corrected P-values (to control FDR for multiple comparisons) for the enrichment of certain cell types (vertical axis), and protein modules (horizontal axis) indicated by module number and color. Significance is demonstrated by the color scales, which go from 0 (white) to 3 (red), representing Àlog(P). Asterisks represent the level of significance of comparisons (*P < 0.05; ***P < 0.005).

and FTD (Fecto & Siddique, 2011), correlated with pathological underlying molecular continuum (Fig 2B). There was an enrich- TDP-43 scores and total TDP-43 levels. Additionally, M7, enriched ment of differentially expressed proteins altered in pairwise with proteins involved in protein transport, correlated with TDP-43 comparisons between ALS/FTD and FTD patients and controls pathology (Fig 3A). that were negatively correlated to neuronal modules (M1 and As expected for the frontal cortex of ALS cases, where TDP-43 M8) and positively correlated to M2, M6, M9, and M10 modules. pathology is typically not found, and no cognitive decline pheno- Additionally, the proteomic network generated in this study was type exists, module eigenprotein expression for ALS patients with- compared to a previously generated proteomic network created out dementia was similar to controls. There was little overlap from frontal cortex samples of control, ALS, AD, and PD cases between controls and either FTD or ALS/FTD. Notably, we found from the Emory Brain Bank (Seyfried et al, 2017). Protein differences in eigenprotein expression between ALS/FTD and FTD modules in both networks were highly preserved (Fig EV3A), and alone; namely, the eigenprotein expression for many of the over-representation analysis revealed that all 15 modules gener- modules representing the ALS/FTD group was intermediate ated in the current study had at least one cognate module within between that of the strictly ALS and strictly FTD clinical groups, the previously generated Emory brain protein network supporting the existence of a molecular spectrum of disease. (Fig EV3B). The consistency between these independent datasets Mapping of differentially expressed proteins to co-expression provides confidence in the networks generated in this study, and modules in cases with ALS/FTD and FTD further supports an an opportunity to expose meaningful relationships along the

52 EMBO Molecular Medicine Vol 10 |No1 | 2018 ª 2017 The Authors Mfon E Umoh et al ALS-FTD proteomic network EMBO Molecular Medicine

ALS-FTD disease spectrum by relating clinical and pathological C-terminal peptide accumulation of TDP-43 is associated with poten- sample traits to modules. tial toxicity of pathological TDP-43 (Zhang et al, 2009). Indeed, we observed an increase in C-terminal TDP-43 when analyzing peptide Protein co-expression classifies the proteome into modules level differences (Fig EV2). Thus, the correlation of inflammatory associated with brain cell types proteins and RNA binding proteins in the M6 and M2 modules, respectively, to clinical and pathological traits, may represent a molec- Co-expression organizes the proteome by placing proteins as nodes ular pathophysiological connection between TDP-43 proteinopathy, within a network with edges that define the connectivity, correla- related post-translational modifications, and consequent changes in tion, of nodes to one another (Gibbs et al, 2013). Many biological the brain proteome occurring along the ALS-FTD disease spectrum. networks have an inherent hierarchical structure that allows nodes to be organized into a small number of highly interconnected Modules associated with inflammation are enriched with TDP-43 modules (Parikshak et al, 2015). Intramodular connectivity creates interactors and have causal links to ALS a rank for proteins within a module, which allows identification of proteins that are module hubs; these are enriched for potential key Modules associated with TDP-43 pathology and clinical phenotypes drivers (Parikshak et al, 2015). Several modules were enriched for described above could either play a causal role in FTD or be microglial and astrocytic proteins (M2, M10, M6, M5, and M13), secondary to the disease process. While postmortem protein expres- while others were enriched for neuronal proteins (M1 and M8), or sion does not directly assess causality of these modules by them- oligodendrocyte-specific proteins (M12 and M4) (Fig 3B), consistent selves, integrating multiple “omic” data sources can prioritize those with previous studies (Oldham et al, 2008; Miller et al, 2013; modules that are most central to FTD and ALS pathogenesis. One Seyfried et al, 2017). Modules of co-expressed proteins enrich with such approach for module prioritization is to assess the enrichment markers of specific brain cell types, and so changes in protein co- of known TDP-43 protein–protein interactions (PPIs) within each expression that relate to clinical phenotypes along the ALS-FTD respective module generated across the ALS-FTD spectrum of cases. disease spectrum may reflect changes in abundance, and/or activa- For example, the M2 and M6 modules were significantly enriched tion of specific cell types (Seyfried et al, 2017). Module gene ontol- for proteins previously identified (Freibaum et al, 2010) in studies ogy enrichment also typically mirrors the cell-type enrichment exploring TDP-43 PPIs (Fig 4A). Furthermore, the protein products analysis results (Dataset EV3). For example, the M6 module was for several genes that have been causally linked to ALS (Taylor enriched for proteins involved in responding to biotic stimuli consis- et al, 2016) are found within the M2 module, including HNRNPA1, tent with the cell-type enrichment of astrocytic and microglial MATR3, and PFN1 and TDP-43 itself (Fig 4B), whereas HSPB1 (in proteins with known roles in neuroinflammation. Modules enriched M6) has been linked to hereditary motor neuropathy, a form of for gene ontologies associated with “synapse-part (GO:0044456)” motor neuron disease (James & Talbot, 2006; Rossor et al, 2011). and “neuronal differentiation (GO:0030182)”, M1 and M8, respec- Additionally, the M8 module, enriched with synaptic proteins, tively, are also enriched for neuronal-specific proteins. Microglial showed enrichment with TDP-43 PPIs, yet lacked any genes causally and astrocytic modules (M2, M6, M5, and M13) were positively linked to ALS or FTD. The strong enrichment of TDP-43 interactors correlated to pathological TDP-43 rating and clinical grouping, while and other causal genes for ALS within M2 and M6 modules further neuronal modules (M1 and M8) were negatively correlated to these reinforces their association with the ALS/FTD spectrum and traits (Fig 3A). M4, enriched with oligodendrocyte proteins, also suggests that proteins within these modules have critical roles in correlated moderately to ALS-FTD clinical grouping (Fig 3A). mechanisms that drive TDP-43 aggregation and other cellular-driven Together, this suggests cell-type-specific processes undergo changes pathological processes (i.e., neuroinflammation). in ALS/FTD that manifest as clinical and pathological endophenotypes, some of which are similar to changes occurring with AD Astrocyte and microglial markers associated with (De Strooper & Karran, 2016; Seyfried et al, 2017). neuroinflammation are increased across the ALS-FTD spectrum

Module relevance to TDP-43 pathology and clinical grouping The M6 module was significantly enriched for TDP-43 PPIs, and positively correlated to pathological TDP-43 burden and clinical The M2 and M6 modules, enriched for RNA binding proteins and dementia. This module includes hepatic and glial cell adhesion inflammatory proteins, respectively, were strongly positively corre- molecule (HEPACAM), membrane-organizing extension spike lated to TDP-43 pathological burden, total TDP-43 levels and clinical protein moesin (MSN), ezrin (EZR), glial fibrillary acidic protein grouping. TDP-43 mapped to M2, a module of co-expressed proteins (GFAP), and peroxiredoxin 6 (PRDX6). HEPACAM, an astrocyte- increased significantly in ALS/FTD and FTD compared to controls. specific protein (Zhang et al, 2014), is involved in regulating cell Notably, this module also included significant enrichment of micro- motility and cell-matrix interactions, while moesin and ezrin are glial proteins (hypergeometric overlap P < 0.001), suggesting a strong members of a family of proteins which function as cross linkers co-expression between RNA binding protein dysfunction and micro- between plasma membranes and actin-based cytoskeleton, also glial inflammation in the ALS-FTD spectrum that was not apparent for regulating motility (Ivetic & Ridley, 2004). Many of the proteins in other neurodegenerative cohorts (Seyfried et al, 2017). As expected, the M6 module are expressed in glial cell types (astrocytes and immunohistochemical analysis of pTDP pathology showed a signifi- microglia), and overrepresent members of the gene ontology cant increase in FTD cases compared to ALS and control cases, which “response to biotic stimuli,” thus defining what we characterize as mirrors overall TDP-43 protein abundance in the frontal cortex of the inflammatory module (Fig 5A). Module eigenprotein expression these brains (Fig EV2). Previous work has suggested that the revealed M6 as a module increased in FTD compared to controls

ª 2017 The Authors EMBO Molecular Medicine Vol 10 |No1 | 2018 53 EMBO Molecular Medicine ALS-FTD proteomic network Mfon E Umoh et al

3 −log(p), BH Corrected

TDP-43 PPIs 1 1 0.39 1 0.00021 1 3.5e−05 1 1 1 1 0.79 1 1 0.067 (BioGRID) 2

1 TDP-43 PPIs 1 1 0.31 0.95 3.5e−05 1 0.0012 1 1 1 1 0.66 1 1 0.0084 (Freibaum et al., 2010)

M7 M11 M15 M9 M2 M10 M6 M5 M13 M3 M1 M12 M4 M14 M8

MYO6 B AK3 ALS GENES

A2M TARDBP MATR3 CAT HDGFRP3

ANK2 NONO HNRNPA1 PFN1 GANAB UGP2 PC CANX BROX S100A11 MT1F BANF1 PRKDC PON2 LTA4H

NPC2 EEF1D HSPB1 PTGR2 BLVRB RPS6KA2 GLUD1 CBX3 PBXIP1 HNRNPF ALDH3A2 STXBP3 MLC1 ITGB1 PNP GYG1 HMGB2 HSDL2 IDH1 ANKFY1 S100A6 GM2A PPAP2B APRT ANXA5 ASAH1 SRI METTL7A DARS PPFIA1 ARPC5 PSME1 CAPN2 NAMPT TSN TROVE2 H2AFY MAOB SEPT10 HSD17B4 PGLS G6PD PYGB TPP1 ADIRF PARK7 CAPG NAGK CD99 GNAO1 GJA1 HIBCH GNG12 LMNB2 GAPDH MT2A AK1 LAMP1 HP1BP3 VPS13C ESD SORD ADH5 FABP5 LHPP CAPN1 AKR1A1 ADD3 HNRNPH3 AQP4 PIR PLCD1 HEPACAM COMT FTH1 RPL23 HSPA1B EZR AHNAK SEPT2 CAPNS1 DDAH2 H2AFV MYO18A PLEC GFAP PSAT1 HNRNPH1 RPL22 RAP1B RAB21 C11orf54 ALDH9A1 TST HSPB1 PGD SLC9A3R1 ACAA2 ERLIN2 CARHSP1 PRDX6 CRYL1 ALAD PGAM2 AGT PTBP1 CNN3 MSN TLN1 HIST2H2AB EIF5A PPA1 HSPB6 PRDX1 HADHB DDX39B GNG10 HRSP12 SPTAN1 PKM VAT1 HNRNPR DHX9 GNA13 EEF1A1 RNH1 APEX1 IQGAP1 GMPR DBI TPPP3 LIN7C PHPT1 COTL1 S100B CBR3 ARL3 MPST GSTP1 CD44 MARCKSL1 SCARB2 S100A16 PADI2 LIMCH1 PHGDH HADHA GPRC5B DDX3X HNRNPUL2 TNS1 AKR7A2 HNRNPK AKR1B1 HIST1H2AC HIST1H2BM CNDP2 NUDT5 BDH2 CSRP1 TKT NPM1 BCLAF1 C4B LMNA ALDH7A1 PPIF SH3BGRL TJP2 LDHB SUCLG2 GSN SBDS MCAM ALDH2 PAICS ALDH1A1 HMGN3 PPIA MTHFD1 HNRNPA2B1 PFN1 FGF1 APOD DDX17 PGM1 ILF2 SRSF7 ALDH6A1 MT1E LLGL1 HIST2H2BE DYNC1LI2 XRCC5 GALK1 EPB41L2 PTGDS YBX1 SNRPD3 HMGB1 CAP1 SSB HNRNPU CA2 PEBP1 KTN1 XRCC6 TJP1 KDSR PRKAR1A GRHPR DYNC1I2 PSMD9 SNX5 CD82 IST1 SCP2 NCL HDHD2 MATR3 SCRN1 MAP4 RTCB HMGN2 PTMA RPE TARDBP TUBB2B KIF5B PLIN3 OSTF1 RPL14 SLC25A1 S100A1 PARP1 HNRNPC MECP2 MAPRE1 RPS28 ALCAM DDX1 NUMA1 HDGF HNRNPA1 ELAVL1 RALY PPIB BOLA2 CDKN1B MEAF6 HNRNPM PCBP1 UBE2L3 PEPD BPNT1 ST13 PRCP EEF2 ILF3 PSMD10 HBD SAR1B ECI1 RRBP1 ENOPH1 TRIM28 HNMT AK2 ANP32A GNAQ AHCY ERH SORT1 CCAR2 HIP1R C7orf55-LUC7L2 EIF3B SRP9 RAB6A GCLC TMEM109 IMPA1 RBMX KHSRP RPS3A DAK HMGA1

FIBP

RPL12

RPLP0

INPP1

RPL11

Figure 4. Modules with causal links to ALS are associated with inflammation and enriched with TDP-43 protein–protein interactions (PPIs). A Enrichment of TDP-43 PPIs is displayed in this heatmap with results from TDP-43 PPI from BIOGRID and those from published global analysis of TDP-43 interacting proteins (Freibaum et al, 2010), with protein co-expression modules listed across the horizontal axis. Colors on heat map indicate enrichment (red) or no significant over-representation (white) for gene membership. Numbers displayed on the heatmaps represent positive signed Àlog10(BH-adjusted P-values). BM2 (blue nodes) and M6 (red nodes) modules represented by proteins with high co-expression (gray lines) and TDP-43 PPIs (yellow lines). Several genes that have been previously linked to ALS-FTD are also highlighted by their colored module membership (inset).

and ALS. The WGCNA measure of intramodular connectivity (kME), the extent to which individual proteins mirror this pattern (Seidel

defined as the Pearson correlation between the expression pattern of et al, 2017). High relative kME can be used to identify individual a protein and the module eigenprotein (which summarizes the char- proteins that best represent a module; typically, these hubs of acteristic expression pattern of proteins within a module), quantifies modules are markers of predominant cell types (Parikshak et al,

54 EMBO Molecular Medicine Vol 10 |No1 | 2018 ª 2017 The Authors Mfon E Umoh et al ALS-FTD proteomic network EMBO Molecular Medicine

2015). Two astrocytic proteins (GFAP, HEPACAM) and two micro- group. Few proteins were differentially expressed in the frontal glial proteins (MSN and TPP1) with high kME were confirmed to be cortex of C9Pos ALS cases compared to C9Neg ALS cases (Dataset increased in abundance in frontal cortex homogenates from FTD EV2 and Fig 6A). Furthermore, none of these C9orf72 genotype- cases compared to that of control and ALS cases by Western blot affected proteins were changed by greater than twofold, making it analysis (Figs 5B and EV4A and B). To assess cellular localization, a difficult to resolve proteome-wide differences by differential expres- representative ALS and ALS/FTD case was immunostained with sion alone. The presence of the C9orf72 expansion as a case-sample antibodies against GFAP, HEPACAM, and MSN and TPP1, which trait also did not significantly correlate with any of the identified displayed astrocytic and microglial cellular populations, respec- protein co-expression module eigengenes in the network when tively, consistent with the specific cell-type expression of these comparisons were drawn across all clinical groups. Together, these protein markers (Fig EV4C). Although qualitative, the relative findings were unexpected, as we anticipated genetic contributions increase in tissue immunoreactivity seen for GFAP, HEPACAM, would generate a clear proteomic signature. However, mapping of MSN, and TPP1 in the ALS/FTD case likely reflects changes in the the differentially expressed proteins in C9Pos ALS to the same abundance of astrocytes and microglia measured by mass spectrom- modules from the co-expression network revealed that those etry and confirmed by Western blotting. Thus, we found increased proteins with significant, yet marginal, fold change in C9Pos ALS inflammation within the frontal cortex in FTD cases, but not ALS, cases, generally mapped into the astrocytic/microglial modules which correlates with TDP-43 abundance, suggesting a potential (positively correlated with gene expansion) and neuronal modules role of glial cells in disease pathogenesis and progression related to (negatively correlated), consistent with changes associated with TDP proteinopathy. cognitive dysfunction in FTD (Fig 6B). Furthermore, an analysis, focused on strictly comparing only the non-demented ALS patients Co-expression defines C9orf72 specific changes in ALS brain with and without the C9orf72 expansion, identified one module related to neuroinflammation eigenprotein (M6) (P < 0.05) as increased in C9Pos ALS (Fig 6C). Notably, these patients were not demented prior to death, suggesting To elucidate the contribution of the C9orf72 genetic expansion to that the C9orf72 mutation may be associated with neuroinflammation disease, we compared protein expression relative to our C9Pos case in the brain. These results also highlight the ability of our integrative

A B Astrocyte Microglia Astrocyte Microglia HEPACAM MSN SPTAN1 MLC1 TNS1 1.0 0.4 ANKFY1 0.5 0.2 PTGR2 H2AFYESD CD99 ANXA5 PGLS PTBP1 MAOB 0.0 Expression PARK7 S100A6 PNP Expression 0.0 2 CAPN2 2 BANF1 AK1 -0.5 ADH5 PLCD1 -0.4 TPP1 TLN1 SH3BGRL

S100A11 SEPT2 PLEC METTL7A CLU PRDX6 ALS FTD ALS FTD ALDH9A1 MSN GNG12 Control Control ALS/FTD ALS/FTD ITGB1 HEPACAM GJA1 EZR PC GFAP TPP1 NAMPT CAPNS1 MT2A 2.0 ADD3 AHNAK 1.0 PGM1 HRSP12 LAMA2 1.0 PON2 GFAP HSPB1 C4B 0.5 IQGAP1 CAPG PBXIP1 0.0 Expression log SRI LDHB Expression log 0.0 2 2 F3 -1.0 ACAA2 log ADIRF log IDH1 PHPT1 -0.5 HSPB6 BDH2 -2.0 LTA4H HP1BP3 PSME1FABP5 HSDL2 ALS FTD ALS FTD HIST1H1E H1F0 Control Control MT1F ALS/FTD ALS/FTD

Figure 5. Astrocyte and microglial markers associated with neuroinflammation are increased across the ALS-FTD spectrum. A I-graph of M6 module (red) representing hub proteins and corresponding gene symbols as nodes. Node size and edges (gray) are reflective of the degree of

intramodular connectivity (kME). Specific cell-type expression is demonstrated by the node rim and text color around the gene symbols (blue: microglial marker and green: astrocytic marker). B Relative abundance measured by label-free quantification (log2 expression values) for the astrocytic hub proteins HEPACAM and GFAP, and the microglial hub proteins MSN and TPP1 were increased in ALS/FTD and FTD cases consistent with the positive correlation of M6 to pathological TDP-43 pathological burden and clinical dementia. Box plots depict mean (horizontal bars) and variance (25th to 75th percentiles), with whiskers extending to the last non-outlier measurements, as shown.

ª 2017 The Authors EMBO Molecular Medicine Vol 10 |No1 | 2018 55 EMBO Molecular Medicine ALS-FTD proteomic network Mfon E Umoh et al

systems approach to resolve genetic vulnerability, which was not and co-expression analyses, we demonstrate that protein expression confidently recognized through differential expression analysis alone. within the brain has a signature that defines the molecular and Although not significant (P = 0.13), we did observe a reduction in the genetic underpinnings of the clinical and pathological ALS-FTD 55-kDa-long isoform C9orf72 protein in C9Pos ALS cases compared to disease spectrum. WGCNA resolved related protein co-expression C9Neg ALS cases consistent with a loss of C9orf72 function patterns or modules representing pathways and brain cell types, and (Fig EV5). This is of interest because haploinsufficiency of C9orf72 showed a decrease in expression levels for modules associated with protein, possibly related to neuroinflammation, has been suggested neurons and an increase in expression of astroglial and microglial as a possible disease mechanism in C9Pos patients (Farg et al,2014; modules associated with cognitive dysfunction and TDP-43 pathol- Waite et al,2014;Yanget al,2016). ogy in brain. Strikingly, C9orf72 expansion in patients dying with ALS predisposed the frontal cortex to elevated co-expressed markers of neuroinflammation, the same ones co-expressed in a concerted Discussion increased pattern across FTD and ALS/FTD patients. Thus, our findings in the context of genetic underpinnings of ALS/FTD and prior Using an unbiased proteomic screen of the postmortem frontal protein co-expression networks for AD (Seyfried et al, 2017) indi- cortex, we were able to identify and quantify protein differences cate that neuroinflammation is a common signature of dementia, along the ALS-FTD disease spectrum. Employing both differential while broad RNA binding protein co-accumulation is more

A B

PRKCB SNRPD1 3.0 ALS: C9pos/C9neg OSBPL1A CCDC6 RPLP1 APEH DYNLT1 KIF21A 0.60 RAN STXBP3 CEP170B DHX9 DSTN

PRPSAP2 Avg. Module Fold Change 2.0 PTGR2 1.0 log VAC14 DMD 2 pvalue 0.40 (C9pos/C9neg) 2 TK2 PBXIP1 U2AF1L4 0.5 -log Neuronal 1.0 Astrocytic & Microglial 0.20 0.0 Differentially Expressed Fraction Module Members -0.5 0.0 0.00 −1.0 −0.5 0.0 0.5 1.0 711159210651331124148

log2 (C9pos/C9neg) Protein Modules

C M6 p = 0.031 M10 p = 0.19 M2 p = 0.18 M8 p = 0.19

0.0 0.1 0.2 0.0 -0.05 0.0 -0.1 0.0 -0.1 -0.1 Eigenprotein -0.15 -0.2 -0.2 -0.2 C9neg C9pos C9neg C9pos C9neg C9pos C9neg C9pos

Figure 6. Co-expression analysis resolves C9orf72-specific changes in ALS brain.

A Volcano plot displays the protein abundance (log2 fold change) against the t-statistic (Àlog10(P-value)) for C9Pos (n = 8) against C9Neg (n = 11) ALS cases. Red and blue dots indicate significantly altered proteins that are increased and decreased, respectively, in C9orf72 carriers. Gray dots below the hashed line represent proteins that are not significantly changed (P > 0.05). Notably, there are no significant proteins that are twofold or more (Ælog2 1.0) increased or decreased in C9Pos cases compared to C9neg cases. B Stacked bar plot represents the analysis of differential expression (pairwise comparison between C9Pos and C9Neg ALS cases) and enrichment of differentially expressed proteins within co-expression modules. Modules are listed along the x-axis, and the height of the bar along the y-axis indicates the proportion of significantly differentially expressed module members, while the color indicates the fold change according to the scale shown. C Module eigenprotein (the summary expression profile of a module) for 4 of the 15 modules generated by comparing C9Pos ALS and C9Neg ALS cases. The M6 module is significant, whereas other glial modules (M10 and M2) and a synaptic module (M8) are increased and decreased in C9Pos cases, respectively, yet do not reach significance using a Student’s t-test (P-value listed above the plot). Box plots depict mean (horizontal bars) and variance (25th to 75th percentiles), with whiskers extending to the last non-outlier measurements, as shown.

56 EMBO Molecular Medicine Vol 10 |No1 | 2018 ª 2017 The Authors Mfon E Umoh et al ALS-FTD proteomic network EMBO Molecular Medicine

pronounced in the current network organized around frontal cortex include additional C9orf72 expansion-positive cases from ALS/FTD of cases on the ALS-FTD clinical spectrum. and pure FTD cases will be critical at resolving genetic drivers of The protein co-expression network generated in this study was disease. Of interest, though, is our finding of increased microglia- consistent with those previously reported in AD (Seyfried et al, and astrocyte-specific proteins in the frontal cortex of C9Pos ALS 2017), which revealed biologically relevant modules linked to speci- patients with no clinically detectable dementia. fic cell types and organelles (e.g., mitochondria). Although the Our results demonstrate the utility of a systems biology approach genetic and pathological drivers in the ALS-FTD spectrum are in understanding complex diseases with underlying comorbidity. distinct from AD, we observed a consistent downregulation of We were able to relate a proteomic signature of disease (i.e., proteo- modules associated with neurons and synapses and upregulation of type) to clinicopathological phenotypes and C9orf72 genotype, glial (microglial and astroglial) modules with increased TDP-43 which has not previously been done. Proteins within co-expression pathology and cognitive dysfunction in brain. This suggests signa- modules could further serve as potential biomarkers of disease tures reflecting relative changes in cellular phenotypes and/or abun- mechanism or targets to assess therapeutic interventions across the dance are shared across neurodegenerative diseases (De Strooper & ALS-FTD disease spectrum. Ultimately, this proteomic study Karran, 2016). However, one clear distinction between the AD provides a resource that moves towards a broad and comprehensive network and our current ALS-FTD network was the definition of the understanding of specific pathways, including inflammation, and M2 module that was enriched with both RNA binding proteins and cell-type-driven processes that relate to the clinical and genetic microglial markers. This contrasts with AD networks in which the underpinnings along the ALS-FTD disease spectrum. microglial and RNA binding proteins segregated to distinct modules (Seyfried et al, 2017). M2 showed a strong correlation with clinical and pathological traits of cases on the ALS/FTD spectrum, co- Materials and Methods expressed with protein products of genes with causative links to ALS (hnRNPs, matrin 3, profilin 1, and TDP-43) and had significant Case details enrichment of TDP-43 PPIs. The strong enrichment of TDP-43 interactors and other causal genes for ALS within this module further All brain tissues were obtained from the Emory Alzheimer’s Disease supports a strong association of M2 with the ALS/FTD spectrum Research Center (ADRC) Brain Bank. Human postmortem tissues and implicates other novel members of this module as having criti- were acquired under proper Institutional Review Board (IRB) proto- cal roles in TDP-43 biology that potentially influence TDP-43 aggre- cols with informed consent from the subject or the family of the gation or other pathological processes (i.e., microglial-directed deceased subject, and all experiments conformed to the principles neuroinflammation) inherent to ALS and FTD etiology. set by the WNA Declaration of Helsinki and the Department of Recent findings have demonstrated links between the C9orf72 Health and Human Services Belmont Report. Cases were selected expansion mutation and inflammatory pathways in C9orf72 animal based on clinical diagnoses. FTD cases were also selected based on models (Burberry et al, 2016; O’Rourke et al, 2016; Sudria-Lopez neuropathological diagnoses to exclude FTD subtypes that did not et al, 2016). Evidence also exists that C9orf72 mutation carriers have neuropathological diagnosis of FTLD-TDP (frontotemporal have an increased prevalence of certain autoimmune disorders lobar degeneration characterized by ubiquitin and TDP-43-positive, (O’Rourke et al, 2016), increased microglial pathology (Brettschnei- tau-negative, FUS-negative inclusions) pathology. Emory neurolo- der et al, 2012), and increased thinning in frontal and temporal gists cared for these patients throughout their disease course, and lobes in neuroimaging studies, compared to sporadic ALS and diagnoses were made using established clinical criteria (Brooks, controls (Floeter et al, 2016). Increased abundance of proteins iden- 1994; Brooks et al, 2000; McKhann et al, 2001; Rascovsky et al, tified within the inflammatory co-expression module in the frontal 2011). Standard diagnostic neuropathological analysis was cortex of ALS/FTD, FTD, and C9Pos ALS cases may explain the clin- performed for all cases, including phosphorylated TDP-43 (pTDP- ical link between cognitive decline and the C9orf72 expansion 43) immunohistochemistry on paraffin-embedded tissue sections (Irwin et al, 2013; Umoh et al, 2016). Thus, from an unbiased using the phosphorylated Ser409/410 antibody (Cosmo Bio). Ordi- proteomic approach, these data support the hypothesis that the nal scales were used to assess pTDP-43 pathology (0–3) with higher dementia phenotype represents changes within the brain that corre- scores indicating greater pathology. The presence of a C9orf72 late with, and are potentially causally linked to, increased neuroin- repeat expansion was assessed from blood samples using the flammation (Bettcher & Kramer, 2013). published repeat primed PCR method (DeJesus-Hernandez et al, There were several limitations in this work. First was the cover- 2011). Clinical and pathological information from all cases, includ- age of the proteome that we were able to achieve with our “single- ing disease status, neuropathological criteria, age, sex, and post- shot” LC-MS/MS approach. The number of proteins identified was mortem interval are provided in Dataset EV1. Notably, 6 of 19 ALS similar to proteomic coverage seen in other studies using similar cases and 1 of 10 control cases processed in this study overlapped approaches (Bi et al, 2017; Seyfried et al, 2017). However, this is with our previous study (Seyfried et al, 2017). much lower than studies investigating the transcriptome. Neverthe- less, the co-expression patterns measured were robust, repro- Homogenization and proteolytic digestion of frontal ducible, and correlated with clinicopathological phenotypes. cortex tissue Another limitation was the relatively small number of FTD cases carrying the C9orf72 expansion; this diminished our power to Dorsolateral prefrontal cortex (Brodmann area 9) tissue samples discern any differences we may have seen due to this genetic muta- were processed essentially as previously described (Seyfried et al, tion between ALS and FTD clinical groups. Future studies that 2017). In brief, each piece of tissue was individually weighed

ª 2017 The Authors EMBO Molecular Medicine Vol 10 |No1 | 2018 57 EMBO Molecular Medicine ALS-FTD proteomic network Mfon E Umoh et al

(~100 mg) and homogenized in 500 ll of urea lysis buffer (8 M modification (+57.0215 Da). Only fully tryptic peptides were consid-

urea, 100 mM NaHPO4, pH 8.5), including 5 ll (100× stock) HALT ered with up to two miscleavages in the database search. A precur- protease and phosphatase inhibitor cocktail (Thermo Fisher, Catalog sor mass tolerance of Æ10 ppm was applied prior to mass accuracy #78440). Homogenization was performed using a Bullet Blender calibration and Æ4.5 ppm after internal MaxQuant calibration. (Next Advance) following manufacturer’s protocols. Each tissue Other search settings included a maximum peptide mass of piece was added to the urea lysis buffer in a 1.5-ml Rino tube (Next 6,000 Da, a minimum peptide length of six residues, 0.6 Da toler- Advance) that contained 750-mg stainless steel beads (0.9–2mmin ance for ion trap HCD MS/MS scans. The false discovery rate (FDR) diameter) and blended twice for 5-min intervals in a 4°C cold room. for peptide spectral matches, proteins, and site decoy fraction were Protein supernatants were transferred into 1.5-ml Eppendorf tubes all set to 1%. The label-free quantitation (LFQ) algorithm in and sonicated (Sonic Dismembrator, Fisher Scientific) for three MaxQuant (Luber et al, 2010; Cox et al, 2014) was used for protein cycles of 5 s (at 30% amplitude) with 15-s intervals of rest to shear quantitation as previously described. To account for possible DNA. Samples were then centrifuged for 2 min at 15,871 g at 4°C. confounds in run time, a brain peptide standard, generated from Protein concentration was assessed using the bicinchoninic acid pooled samples of homogenized brain, was included at different (BCA) method, and samples were frozen at À80°C until use. Each points in the run set to control for drift over time and highlight homogenate was analyzed by SDS–PAGE to assess protein integrity. consistency in the protein measurements (Fig EV1A). Brain protein homogenates (100 lg) were diluted with 50 mM

NH4HCO3 to a final concentration of < 2 M urea. Samples were Data analysis and pre-processing subsequently treated with 1 mM dithiothreitol (DTT) at 25°C for 30 min and then 5 mM iodoacetamide (Engelhart et al, 2004) at Protein filtering and data imputation 25°C for 30 min in the dark. Protein samples were digested with Protein abundance was determined by peptide ion-intensity 1:100 (w:w) lysyl endopeptidase (Wako) at 25°C for 2 h and then measurements across LC-MS runs using the label-free quantification further digested with 1:50 (w/w) trypsin (Promega) overnight at (LFQ) algorithm in MaxQuant (Cox et al, 2014). In total, 47,977 25°C. Resulting peptides were desalted with a Sep-Pak C18 column peptides mapping to 4,178 protein groups were identified. However, (Waters) and dried under vacuum. one limitation of data-dependent label-free quantitative proteomics is missing quantitative measures, especially for low-abundance Liquid chromatography coupled to tandem mass proteins (Karpievitch et al, 2012; Seyfried et al, 2017). Thus, only spectrometry (LC-MS/MS) those proteins quantified in at least 90% of samples were included in the data analysis. After filtering, only allowing 10% missing values For LC-MS/MS analysis, the peptides were first resuspended in maximum across the 51 LC-MS/MS runs, 2,612 unique proteins were 100 ll of loading buffer (0.1% formic acid, 0.03% trifluoroacetic identified and robustly quantified (Fig EV1B). The 10% or fewer acid, 1% acetonitrile). Peptide mixtures (2 ll) were separated on a missing protein LFQ values were imputed using the k-nearest neigh- self-packed C18 (1.9 lm Dr. Maisch, Germany) fused silica column bor imputation function in R impute::impute.knn() function, similar (25 cm × 75 lM internal diameter (ID); New Objective, Woburn, to what has been previously described (Seyfried et al, 2017). MA, USA) by a Dionex Ultimate 3000 RSLCNano and monitored on a Fusion mass spectrometer (ThermoFisher Scientific, San Jose, CA, Outlier removal and regression USA). Elution was performed over a 140-min total gradient at a rate Prior to data analysis, outlier removal was performed using of 300 nl/min with buffer B ranging from 1 to 65% (buffer A: 0.1% Oldham’s “SampleNetworks” v1.06 R script (Oldham et al, 2008) as formic acid in water, buffer B: 0.1% formic in acetonitrile). The previously published (Seyfried et al, 2017). Two control and two mass spectrometer cycle was programmed to collect at the top speed FTD cases were removed from the 51 cases initially included for 3-s cycles. The MS scans (400–1,600 m/z range, 200,000 AGC, (Fig EV1B). Bootstrap regression of the remaining 47-case LFQ 50 ms maximum ion time) were collected at a resolution of 120,000 intensity matrix that explicitly modeled case status category while at m/z 200 in profile mode and the HCD MS/MS spectra (0.7 m/z removing covariation with age at death, gender, and PMI was done isolation width, 30% collision energy, 10,000 AGC target, 35 ms following principal component analysis (PCA) of the expression maximum ion time) were detected in the ion trap. Dynamic exclu- data to confirm appropriate regression of selected traits, both in the sion was set to exclude previous sequenced precursor ions for 20 s “SampleNetworks” graphical output and via an in-house R script for within a Æ10 ppm window. Precursor ions with +1, and +8 or higher PCA Spearman correlation to the amassed traits for 47 all non- charge states were excluded from sequencing. outlier cases. PCA visualized that the top five principal components had Spearman correlation q < 0.3 with any of these three regressed Label-free protein quantification covariates, and < 0.02 after regression.

Raw data files were analyzed using MaxQuant v1.5.2.8 with Thermo Differential expression analysis Foundation 2.0 for RAW file reading capability essentially as Differentially enriched or depleted proteins (P ≤ 0.05) were identi- described with slight modifications (Seyfried et al, 2017). The fied by ANOVA comparing the four clinical groups (control, ALS, search engine Andromeda was used to build and search a concate- ALS/FTD, and FTD). Multidimensional scaling as implemented in nated target-decoy Uniprot human reference database. Protein the WGCNA R package (Langfelder & Horvath, 2008) was used to methionine oxidation (+15.9949 Da) and protein N-terminal acetyla- visualize separation of cases using a subset of 165 proteins which tion (+42.0106 Da) were variable modifications (up to five allowed had at least two ANOVA Tukey pairwise comparisons of high per peptide); cysteine was assigned a fixed carbamidomethyl significance (P < 0.01) among the six possible comparisons

58 EMBO Molecular Medicine Vol 10 |No1 | 2018 ª 2017 The Authors Mfon E Umoh et al ALS-FTD proteomic network EMBO Molecular Medicine

among ALS, FTD, FTD/ALS, and control groups. In a separate v1.2.5 as previously published (Seyfried et al, 2017), with pruned analysis, differentially expressed proteins in C9orf72 expansion- output visualized using an in-house R script (Dataset EV3). Cell-type positive (C9Pos) ALS versus C9orf72 expansion negative (C9Neg) enrichment was also investigated as previously published (Seyfried ALS (excluding cases with coexisting dementia) were identified et al, 2017). Enrichment of TDP-43 PPI across co-expression by t-test. Differential expression is presented as volcano plots, modules was investigated by intersecting module proteins with lists which were generated with the ggplot2 package in Microsoft R of genes known to interact with TDP-43, and assessing significance Open v3.3.2. Significantly altered proteins along with correspond- of overlap using a one-tailed Fisher exact hypergeometric overlap ing P-value are listed in Dataset EV2. test. TDP-43 PPI lists from BioGRID (https://thebiogrid.org/ 117003/summary/homo-sapiens/tardbp.html), and a previously Co-expression network analysis published global analysis of TDP-43 interacting proteins (Freibaum Following previously described procedures of WGCNA (Seyfried et al, 2010) was used. The total list of identified protein groups was et al, 2017), a weighted protein co-expression network was gener- used as the background and the PPIs lists (Dataset EV4) were fil- ated using this pre-processed protein abundance matrix, using the tered for presence in the total proteins list prior to cross-referencing. WGCNA::blockwiseModules() function with the following settings: After assessing significance of TDP-43 PPIs, P-values were corrected soft threshold power beta = 4.5, deepsplit = 4, minimum module for multiple comparisons by the Benjamini–Hochberg method. size of 12, merge cut height of 0.07, signed network with parti- tioning about medoids (PAM) respecting the dendrogram and a Western blotting reassignment threshold of P < 0.05. Specifically, we calculated pairwise biweight mid-correlations (bicor, a robust correlation metric) Frontal cortex tissue homogenates in Laemmli sample buffer were between each protein pair and transformed that matrix into a signed resolved by SDS–PAGE [NuPAGE Bis-Tris (Life Technologies)]. Gels adjacency matrix (Langfelder & Horvath, 2012). The connection were transferred onto nitrocellulose membranes (Invitrogen) using strength of components within this matrix was used to calculate a the iBlot 7-min dry transfer blotting system (Thermo Fisher Scien- topological overlap matrix, which represents measurements of tific). Blots were blocked with TBS starting block buffer (Thermo protein expression pattern similarity across the set of samples in the Fisher Scientific) for 30 min at room temperature and then probed cohort constructed on the pairwise correlations for all proteins with primary antibodies diluted in 10% blocking buffer in PBS over- within the network (Yip & Horvath, 2007). Hierarchical protein night at 4°C. The next day, blots were rinsed and incubated with correlation clustering analysis by this approach was conducted secondary antibodies conjugated to fluorophores, Alexa Fluor680 using 1-TOM, and initial module identifications were established goat anti-mouse IgG(H+L) or Alexa Fluor680 goat anti- Rabbit IgG using dynamic tree cutting as implemented in the WGCNA::block- (H+L) (Life Technologies), for 1 h at room temperature. Images wiseModules() function (Langfelder et al, 2008). Module eigenpro- were captured using an Odyssey Infrared Imaging system (LiCor teins were defined, which represent the most representative Biosciences). Protein densitometry for relative quantification was abundance value for a module and which explain covariance of all performed using ImageJ open source software. Higher molecular proteins within a module (Miller et al, 2013). Pearson correlations weight protein isoforms were considered in the quantification for between each protein and each module eigenprotein were both TPP1 and HEPACAM as each has been shown to be glycosy- performed; this module membership measure is defined as kME. lated (Golabek et al, 2003; Moh et al, 2005). Antibodies used Figure EV1B illustrates workflow for analysis. include TDP-43 (Proteintech, 10782-2-AP; 1:1,000), tripeptidyl peptidase 1 (TPP1) (Sigma-Aldrich, WH0001200M1; 3 lg/ml), GFAP Module preservation and over-representation analyses (Millipore, MAB360; 1:1,000), C9orf72 (Abcam, ab183892; 1:1,000), Module preservation was tested using the “modulePreservation” moesin (MSN) (Abcam, ab50007; 1:2,500), hepatic and glial cell WGCNA R package function, using exactly 500 permutations adhesion molecule (HEPACAM) (Abcam, ab130769; 1:1,000), and comparing the frontal cortex proteomic network generated from glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (EMD Milli- samples in this study against a previously generated Emory human pore, AB2302; 1:1,000) as a loading control. brain protein network, a similar network built with ALS and other neurodegenerative disease cases from the Emory brain bank Immunohistochemistry (Seyfried et al, 2017). This analysis was to ensure that the modules were representative of frontal cortex specific networks. Additionally, Paraffin-embedded sections of frontal cortex (8 lm thickness) were over-representation analysis (ORA) was conducted using Fisher deparaffinized by incubation at 60°C for 30 min and rehydrated by exact tests (two-tailed) between module membership of the ALS- immersion in graded ethanol solutions. Antigen retrieval was FTD proteomic network versus the previous Emory brain protein performed by microwaving slides in 10 mM citrate buffer pH 6.0 for network (Seyfried et al, 2017). ORA analysis uses gene set enrich- 5 min and then allowing slides to cool to room temperature (RT) for ment analysis with a two-tailed Fisher exact test with 95% confi- 30 min. Peroxidase quenching was performed by incubating slides dence intervals by employing the R function “fisher.test” (Seyfried in a 3% hydrogen peroxide solution in methanol for 5 min at 40°C. et al, 2017). To reduce false positives, Benjamini–Hochberg FDR Slides were then rinsed in Tris-Brij buffer (1 M Tris-Cl pH 7.5, adjustment of P-values corrected for multiple comparisons. 100 mM NaCl, 5 mM MgCl2, 0.125% Brij 35). For blocking, sections were incubated in normal goat serum or normal horse serum (Elite Enrichment analyses Vectastain ABC kit), depending on the primary antibody species, for To characterize differentially expressed proteins and co-expressed 15 min at 40°C. Sections were then incubated with primary antibod- proteins based on gene ontology annotation, we used GO Elite ies (diluted in 1% BSA in Tris-brij 7.5) overnight at 4°C. The

ª 2017 The Authors EMBO Molecular Medicine Vol 10 |No1 | 2018 59 EMBO Molecular Medicine ALS-FTD proteomic network Mfon E Umoh et al

UCLA) for helpful comments and discussion. We also thank the clinical teams The paper explained whose analyses and care contributed to the dataset. Support was provided by 01 046161 02 Problem the Accelerating Medicine Partnership AD grant U AG - , the NINDS Amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) Emory Neuroscience Core (P30NS055077), and the Emory Alzheimer’s Disease are distinct neurodegenerative diseases that overlap in clinical presen- Research Center (P50AG025688). N.T.S. is supported by 5R01AG053960-02 and 9 72 tation, neuropathology, and genetic underpinnings (e.g., C orf ), yet in part by an Alzheimer’s Association (ALZ), Alzheimer’s Research UK (ARUK), the molecular basis for this overlap is not well understood. The Michael J. Fox Foundation for Parkinson’s Research (MJFF), and the Weston Brain Institute Biomarkers Across Neurodegenerative Diseases Grant Results 11060 32 We performed the first comparative quantitative mass spectrometry- ( ). M.E.U. was also funded by a pre-doctoral T NINDS training grant based proteomic analysis of human postmortem brain tissues reveal- 3T32NS007480-16. ing molecular signatures distinguishing clinicopathological phenotypes across the ALS-FTD spectrum of disease. A systems-level co-expression Author contributions network approach generated modules or groups of highly correlated MEU: Designed and performed experiments, analyzed the data, and wrote proteins associated with RNA binding, synaptic transmission, inflammation, and specific cell types (neuronal, microglial, and astro- original and final draft of manuscript. EBD: Designed, implemented, and cytic), which showed strong correlation with neuropathology and performed computational analyses, and participated in manuscript drafting cognitive dysfunction. Modules were also examined for their overlap and editing. JD: Performed experiments and analyzed the data. DMD: Designed, 43 with TDP- interacting proteins, revealing one module enriched with implemented, and performed experimental analyses. JJL: Participated in manu- RNA binding proteins and other causal ALS genes. Compared to script drafting and editing. AIL: Participated in manuscript drafting and editing. sporadic ALS patients, C9orf72-positive ALS cases showed increased levels for a protein module associated with astrocytic and microglial MG: Designed, implemented, and performed experimental analyses. JDG: markers, supporting a hypothesis linking the C9orf72 mutation to Designed and implemented analyses, and participated in manuscript drafting neuroinflammation. and editing. NTS: Designed and implemented analyses, and participated in manuscript drafting and editing. All authors contributed to and have approved Impact the final manuscript. Our systems-level analysis of the brain proteome offers new insights into the pathways and cell types underlying clinical phenotypes across the ALS-FTD spectrum and provides evidence that the C9orf72 muta- Conflict of interest tion is associated with neuroinflammation in brain. The authors declare that they have no conflict of interest.

References following day sections were incubated in biotinylated secondary antibody at 5 ll/ml (Elite IgG Vectastain ABC kit) for 30 min at Achi EY, Rudnicki SA (2012) ALS and frontotemporal dysfunction: a review. 37°C and then incubated with the avidin-biotin enzyme complex Neurol Res Int 2012: 806306 (Vector Laboratories) for 30 min. Stains were visualized by incu- Bettcher BM, Kramer JH (2013) Inflammation and clinical presentation in bation of DAB Chromogen (Sigma-Aldrich) for 5 min at RT. Slides neurodegenerative disease: a volatile relationship. Neurocase 19: 182 – 200 were then dehydrated in an ethanol series and mounted with Bi B, Li F, Guo J, Li C, Jing R, Lv X, Chen X, Wang F, Azadzoi KM, Wang L et al coverslips. Slides were analyzed using an Olympus BX51 micro- (2017) Label-free quantitative proteomics unravels the importance of RNA scope and imaged with an Olympus DP70 camera. Antibodies processing in glioma malignancy. Neuroscience 351: 84 – 95 used include phosphorylated TDP-43 (pTDP-43) (Cosmo Bio 409/ Brettschneider J, Toledo JB, Van Deerlin VM, Elman L, McCluskey L, Lee VM, 410, TIP-PTD-P02; 1:1,000 for IHC), TPP1 (Sigma-Aldrich, Trojanowski JQ (2012) Microglial activation correlates with disease WH0001300M1; 3 lg/ml), GFAP (Millipore, MAB360; 1:1,000), progression and upper motor neuron clinical symptoms in amyotrophic moesin (Abcam, ab50007; 2 lg/ml), and HEPACAM (Abcam, lateral sclerosis. PLoS One 7:e39216 ab130769; 5 lg/ml). Pathological sample traits were represented Brettschneider J, Del Tredici K, Irwin DJ, Grossman M, Robinson JL, Toledo JB, by the abundance of phosphorylated TDP-43 cytoplasmic inclu- Lee EB, Fang L, Van Deerlin VM, Ludolph AC et al (2014) Sequential sions in the frontal cortex (defined by the pTDP score) or label- distribution of pTDP-43 pathology in behavioral variant frontotemporal free quantification of TDP-43 (Fig EV2). dementia (bvFTD). Acta Neuropathol 127: 423 – 439 Brooks BR (1994) El Escorial World Federation of Neurology criteria for the Data availability diagnosis of amyotrophic lateral sclerosis. Subcommittee on Motor Neuron Diseases/Amyotrophic Lateral Sclerosis of the World Federation of All raw data, MaxQuant output files, and analysis code used in this Neurology Research Group on Neuromuscular Diseases and the El Escorial publication are available from Synapse (www.synapse.org) via “Clinical limits of amyotrophic lateral sclerosis” workshop contributors. J accession syn10142580. Neurol Sci 124(Suppl): 96 – 107 Brooks BR, Miller RG, Swash M, Munsat TL (2000) El Escorial revisited: revised Expanded View for this article is available online. criteria for the diagnosis of amyotrophic lateral sclerosis. Amyotroph Lateral Scler Other Motor Neuron Disord 1: 293 – 299 Acknowledgements Burberry A, Suzuki N, Wang JY, Moccia R, Mordes DA, Stewart MH, Suzuki- We are grateful to the patients and families that donated tissue samples to Uematsu S, Ghosh S, Singh A, Merkle FT et al (2016) Loss-of-function the Emory University brain bank for their contributions to this study. We mutations in the C9ORF72 mouse ortholog cause fatal autoimmune acknowledge Dr. Daniel Geschwind and Vivek Swarup (Department of Genetics, disease. Sci Transl Med 8: 347393

60 EMBO Molecular Medicine Vol 10 |No1 | 2018 ª 2017 The Authors Mfon E Umoh et al ALS-FTD proteomic network EMBO Molecular Medicine

Carvey PM, Hendey B, Monahan AJ (2009) The blood brain barrier in Karpievitch YV, Dabney AR, Smith RD (2012) Normalization and missing value neurodegenerative disease: a rhetorical perspective. J Neurochem 111: imputation for label-free LC-MS analysis. BMC Bioinformatics 13:S5 291 – 314 Langfelder P, Horvath S (2008) WGCNA: an R package for weighted Cox J, Hein MY, Luber CA, Paron I, Nagaraj N, Mann M (2014) Accurate correlation network analysis. BMC Bioinformatics 9: 559 proteome-wide label-free quantification by delayed normalization and Langfelder P, Zhang B, Horvath S (2008) Defining clusters from a hierarchical maximal peptide ratio extraction, termed MaxLFQ. Mol Cell Proteomics 13: cluster tree: the Dynamic Tree Cut package for R. Bioinformatics 24: 719 – 720 2513 – 2526 Langfelder P, Horvath S (2012) Fast R functions for robust correlations and De Strooper B, Karran E (2016) The cellular phase of Alzheimer’s disease. Cell hierarchical clustering. J Stat Softw 46:i11 164: 603 – 615 Lomen-Hoerth C (2011) Clinical phenomenology and neuroimaging correlates DeJesus-Hernandez M, Mackenzie IR, Boeve BF, Boxer AL, Baker M, Rutherford in ALS-FTD. J Mol Neurosci 45: 656 – 662 NJ, Nicholson AM, Finch NA, Flynn H, Adamson J et al (2011) Expanded Luber CA, Cox J, Lauterbach H, Fancke B, Selbach M, Tschopp J, Akira S, Wiegand GGGGCC hexanucleotide repeat in noncoding region of C9ORF72 causes M, Hochrein H, O’Keeffe M et al (2010) Quantitative proteomics reveals chromosome 9p-linked FTD and ALS. Neuron 72: 245 – 256 subset-specific viral recognition in dendritic cells. Immunity 32: 279 – 289 Engelhart M, Geerlings M, Meijer J, Kiliaan A, Ruitenberg A, van Swieten J, Majounie E, Renton AE, Mok K, Dopper EGP, Waite A, Rollinson S, Chiò A, Stijnen T, Hofman A, Witteman J, Breteler M (2004) Inflammatory proteins Restagno G, Nicolaou N, Simon-Sanchez J et al (2012) Frequency of the in plasma and the risk of dementia: the rotterdam study. Arch Neurol 61: C9orf72 hexanucleotide repeat expansion in patients with amyotrophic 668 – 672 lateral sclerosis and frontotemporal dementia: a cross-sectional study. Farg MA, Sundaramoorthy V, Sultana JM, Yang S, Atkinson RA, Levina V, Lancet Neurol 11: 323 – 330 Halloran MA, Gleeson PA, Blair IP, Soo KY et al (2014)C9ORF72, implicated McKhann GM, Albert MS, Grossman M, Miller B, Dickson D, Trojanowski JQ in amytrophic lateral sclerosis and frontotemporal dementia, regulates (2001) Clinical and pathological diagnosis of frontotemporal dementia: endosomal trafficking. Hum Mol Genet 23: 3579 – 3595 report of the Work Group on Frontotemporal Dementia and Pick’s Disease. Fecto F, Siddique T (2011) Making connections: pathology and genetics link Arch Neurol 58: 1803 – 1809 amyotrophic lateral sclerosis with frontotemporal lobe dementia. J Mol Miller JA, Oldham MC, Geschwind DH (2008) A systems level analysis of Neurosci 45: 663 – 675 transcriptional changes in Alzheimer’s disease and normal aging. J Ferrari R, Kapogiannis D, Huey ED, Momeni P (2011) FTD and ALS: a tale of Neurosci 28: 1410 – 1420 two diseases. Curr Alzheimer Res 8: 273 – 294 Miller JA, Woltjer RL, Goodenbour JM, Horvath S, Geschwind DH (2013) Genes Floeter MK, Bageac D, Danielian LE, Braun LE, Traynor BJ, Kwan JY (2016) and pathways underlying regional and cell type changes in Alzheimer’s Longitudinal imaging in C9orf72 mutation carriers: relationship to disease. Genome Med 5: 48 phenotype. Neuroimage Clin 12: 1035 – 1043 Moh MC, Zhang C, Luo C, Lee LH, Shen S (2005) Structural and functional Fong JC, Karydas AM, Goldman JS (2012) Genetic counseling for FTD/ALS analyses of a novel Ig-like cell adhesion molecule, hepaCAM, in the caused by the C9ORF72 hexanucleotide expansion. Alzheimers Res Ther 4: human breast carcinoma MCF7 cells. J Biol Chem 280: 27366 – 27374 27 Neumann M, Sampathu DM, Kwong LK, Truax AC, Micsenyi MC, Chou TT, Freibaum BD, Chitta R, High AA, Taylor JP (2010) Global analysis of TDP-43 Bruce J, Schuck T, Grossman M, Clark CM et al (2006) Ubiquitinated TDP- interacting proteins reveals strong association with RNA splicing and 43 in frontotemporal lobar degeneration and amyotrophic lateral sclerosis. translation machinery. J Proteome Res 9: 1104 – 1120 Science 314: 130 – 133 Gaiteri C, Ding Y, French B, Tseng GC, Sibille E (2014) Beyond modules and Neumann M, Rademakers R, Roeber S, Baker M, Kretzschmar HA, Mackenzie hubs: the potential of gene coexpression networks for investigating IRA (2009) A new subtype of frontotemporal lobar degeneration with FUS molecular mechanisms of complex brain disorders. Genes Brain Behav 13: pathology. Brain 132: 2922 – 2931 13 – 24 Oldham MC, Konopka G, Iwamoto K, Langfelder P, Kato T, Horvath S, Gibbs DL, Baratt A, Baric RS, Kawaoka Y, Smith RD, Orwoll ES, Katze MG, Geschwind DH (2008) Functional organization of the transcriptome in McWeeney SK (2013) Protein co-expression network analysis (ProCoNA). J human brain. Nat Neurosci 11: 1271 – 1282 Clin Bioinforma 3: 11 Oldham MC (2014) Transcriptomics: from differential expression to Gitler AD, Tsuiji H (2016) There has been an awakening: emerging coexpression. In The OMICs: applications in neuroscience, Coppola G (ed), mechanisms of C9orf72 mutations in FTD/ALS. Brain Res 1647: 19 – 29 pp 85 – 113. New York, NY: Oxford University Press Golabek AA, Kida E, Walus M, Wujek P, Mehta P, Wisniewski KE (2003) O’Rourke JG, Bogdanik L, Yanez A, Lall D, Wolf AJ, Muhammad AK, Ho R, Biosynthesis, glycosylation, and enzymatic processingin vivo of human Carmona S, Vit JP, Zarrow J et al (2016)C9orf72 is required for proper tripeptidyl-peptidase I. J Biol Chem 278: 7135 – 7145 macrophage and microglial function in mice. Science 351: 1324 – 1329 Golde TE, Borchelt DR, Giasson BI, Lewis J (2013) Thinking laterally about Parikshak NN, Gandal MJ, Geschwind DH (2015) Systems biology and gene neurodegenerative proteinopathies. J Clin Investig 123: 1847 – 1855 networks in neurodevelopmental and neurodegenerative disorders. Nat Irwin DJ, McMillan CT, Brettschneider J, Libon DJ, Powers J, Rascovsky K, Rev Genet 16: 441 – 458 Toledo JB, Boller A, Bekisz J, Chandrasekaran K et al (2013) Cognitive Radford RA, Morsch M, Rayner SL, Cole NJ, Pountney DL, Chung RS (2015) The decline and reduced survival in C9orf72 expansion frontotemporal established and emerging roles of astrocytes and microglia in degeneration and amyotrophic lateral sclerosis. J Neurol Neurosurg amyotrophic lateral sclerosis and frontotemporal dementia. Front Cell Psychiatry 84: 163 – 169 Neurosci 9: 414 Ivetic A, Ridley AJ (2004) Ezrin/radixin/moesin proteins and Rho GTPase Rascovsky K, Hodges JR, Knopman D, Mendez MF, Kramer JH, Neuhaus J, van signalling in leucocytes. Immunology 112: 165 – 176 Swieten JC, Seelaar H, Dopper EG, Onyike CU et al (2011) Sensitivity of James PA, Talbot K (2006) The molecular genetics of non-ALS motor neuron revised diagnostic criteria for the behavioural variant of frontotemporal diseases. Biochim Biophys Acta 1762: 986 – 1000 dementia. Brain 134: 2456 – 2477

ª 2017 The Authors EMBO Molecular Medicine Vol 10 |No1 | 2018 61 EMBO Molecular Medicine ALS-FTD proteomic network Mfon E Umoh et al

Renton AE, Majounie E, Waite A, Simon-Sanchez J, Rollinson S, Gibbs JR, Webster CP, Smith EF, Bauer CS, Moller A, Hautbergue GM, Ferraiuolo L, Schymick JC, Laaksovirta H, van Swieten JC, Myllykangas L et al (2011)A Myszczynska MA, Higginbottom A, Walsh MJ, Whitworth AJ et al (2016) hexanucleotide repeat expansion in C9ORF72 is the cause of chromosome The C9orf72 protein interacts with Rab1a and the ULK1 complex to 9p21-linked ALS-FTD. Neuron 72: 257 – 268 regulate initiation of autophagy. EMBO J 35: 1656 – 1676 Rossor AM, Kalmar B, Greensmith L, Reilly MM (2011) The distal hereditary Yang M, Liang C, Swaminathan K, Herrlinger S, Lai F, Shiekhattar R, Chen JF motor neuropathies. J Neurol Neurosurg Psychiatry 83: 6 (2016)AC9ORF72/SMCR8-containing complex regulates ULK1 and plays a Seidel K, Marangoni P, Tang C, Houshmand B, Du W, Maas RL, Murray S, dual role in autophagy. Sci Adv 2:e1601167 Oldham MC, Klein OD (2017) Resolving stem and progenitor cells in Yip AM, Horvath S (2007) Gene network interconnectedness and the the adult mouse incisor through gene co-expression analysis. Elife 6: generalized topological overlap measure. BMC Bioinformatics 8: 22 e24712 Zhang B, Horvath S (2005) A general framework for weighted gene co- Seyfried NT, Dammer EB, Swarup V, Nandakumar D, Duong DM, Yin L, Deng expression network analysis. Stat Appl Genet Mol Biol 4: Article17 Q, Nguyen T, Hales CM, Wingo T et al (2017) A multi-network approach Zhang YJ, Xu YF, Cook C, Gendron TF, Roettges P, Link CD, Lin WL, Tong J, identifies protein-specific co-expression in asymptomatic and Castanedes-Casey M, Ash P et al (2009) Aberrant cleavage of TDP-43 symptomatic Alzheimer’s disease. Cell Syst 4: 60 – 72.e64 enhances aggregation and cellular toxicity. Proc Natl Acad Sci USA 106: Sudria-Lopez E, Koppers M, de Wit M, van der Meer C, Westeneng HJ, Zundel 7607 – 7612 CA, Youssef SA, Harkema L, de Bruin A, Veldink JH et al (2016) Full ablation Zhang D, Iyer LM, He F, Aravind L (2012) Discovery of novel DENN proteins: of C9orf72 in mice causes immune system-related pathology and neoplastic implications for the evolution of eukaryotic intracellular membrane events but no motor neuron defects. Acta Neuropathol 132: 145 – 147 structures and human disease. Front Genet 3: 283 Taylor JP, Brown RH Jr, Cleveland DW (2016) Decoding ALS: from genes to Zhang Y, Chen K, Sloan SA, Bennett ML, Scholze AR, O’Keeffe S, Phatnani HP, mechanism. Nature 539: 197 – 206 Guarnieri P, Caneda C, Ruderisch N et al (2014) An RNA-sequencing Umoh ME, Fournier C, Li Y, Polak M, Shaw L, Landers JE, Hu W, Gearing M, transcriptome and splicing database of glia, neurons, and vascular cells of Glass JD (2016) Comparative analysis of C9orf72 and sporadic disease in the cerebral cortex. J Neurosci 34: 11929 – 11947 an ALS clinic population. Neurology 87: 1024 – 1030 Waite AJ, Baumer D, East S, Neal J, Morris HR, Ansorge O, Blake DJ (2014) License: This is an open access article under the Reduced C9orf72 protein levels in frontal cortex of amyotrophic terms of the Creative Commons Attribution 4.0 lateral sclerosis and frontotemporal degeneration brain with the License, which permits use, distribution and reproduc- C9ORF72 hexanucleotide repeat expansion. Neurobiol Aging 35: tion in any medium, provided the original work is 1779 .e5–e13 properly cited.

62 EMBO Molecular Medicine Vol 10 |No1 | 2018 ª 2017 The Authors